Implementation of a supervised learning technique in a multi-agent system for building production orders

Abstract

The authors describe the implementation of a supervised learning algorithm within a multi-agent system, whose general objective is to build production orders. Although this task has been carried out traditionally by the production management system, the classic approach lacks adaptive techniques and intelligent behavior. It is acknowledged that the combinatorial problem underlying the construction of production orders belongs to the NP hard complexity class. Therefore, flexible computational solutions are needed. We claim that by using intelligence and collaboration in a multi-agent system (MAS), a correct solution is reached more efficiently. Intelligence is emulated by both learning and decision-making, achieved through a feed-forward artificial neural network (FANN). The FANN is embedded in a machine agent, which determines the appropriate machine to manufacture the product. Collaboration is obtained by employing a sound protocol based on FIPA-ACL messages. We illustrate the approach by designing and implementing a MAS, which is already in use in a company that produces labels.